AI Paper Atlas

Research map / Label-efficient and robust learning / Knowledge distillation

Foundry: Distilling 3D Foundation Models for the Edge

Guillaume Letellier, Siddharth Srivastava, Frederic Jurie, Gaurav Sharma

CVPR 2026 · Label-efficient and robust learning › Knowledge distillation · approach: Few-shot and metric distillation · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
distill one generalist student instead of task-specific specialistsFoundry · ShapeNet55accuracy (after fine-tuning): 75.09* -> 89.95Improved
replace learnable DSO with static K-Means (ablation)Foundry · ShapeNet55accuracy: 89.68 -> 76.08Hurt
use FPS pre-sampling instead of learned compression (ablation)Foundry · ShapeNet55accuracy: 89.68 -> 87.56Hurt
unfreeze student encoder during distillation (ablation)Foundry · Distillation lossdistillation loss: 0.1006 -> 0.0736Improved
soft attention assignment instead of hard argmax (ablation)Foundry · Distillation lossdistillation loss: 0.0736 -> 0.0787Hurt
add QKV biases in DSO (ablation)Foundry · Distillation lossdistillation loss: 0.0736 -> 0.0763Hurt
add output layer norm at CAU (ablation)Foundry · Distillation lossdistillation loss: 0.0736 -> 0.0748Hurt
increase number of SuperTokens beyond 4 (ablation)Foundry · ModelNet40 and othersaccuracy: 91.25 -> 91.41No clear effect
combine FMD distillation with ToMeToMe (s=16) · Avg over benchmarksaverage score: 73.80 -> 86.52Improved
combine FMD distillation with PatchMergerPatchMerger (s=16) · Avg over benchmarksaverage score: 87.59 vs teacher 87.72Improved
compress to a single SuperToken (s=1 versus s=16) (ablation)Foundry · ModelNet40 5-way 10-shotaccuracy: 92.5 -> 91.8No clear effect
increase gate regularization lambda_gate (ablation)Foundry-Gate · ModelNet40Spearman correlation with performance: -0.45 (frozen), -0.13 (unfrozen)Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy